Audio Analyzing with Algorithm help

Hello people, I'm new to this arduino thing and I have an idea for our project in school.. It's a smart trash bin.. example, when I throw in a paper to the bin it will analyze it's sound if it's really paper and push it in to the biodegradable side of the bin and same goes for non bio.. I figured I'd definitely use a sound sensor then pre record some audio on paper and etc. then something to analyze the audio and the realtime sensing...

I don't really know how to explain this but I hope some of you get where I'm going at..

sorry for being such a newb, :sweat_smile:

Yes I get the idea, and it is often asked about in one form or other, baby crying, dog barking and so on.

then something to analyze the audio and the realtime sensing

That is the difficult / impossible thing to do unfortunately, two sounds can sound the same to us but produce vastly different waveforms.

The best option you could try is an Arduino voice recognition shield and see if you can train it to your sounds. But I don’t hold out much hope, a lot depends on how many false positives and negative triggers you can put up with for your application.

You can do it, but not with Arduino. For voice recognition to work like this, you need more advanced chip like Kendryte or X86 Linux / Android smart watch with GRT C++ framework (Github: GRT, spoiler: I help with this project). Or Google Tensorflow.

I am sound engineer and music producer. The waveform complexity is not a problem. Problem is you need better computation engine and Arduino voice recognition shield is not that. Because of: low number of samples in a dataset, optimized for voice (not good frequency spectrum), cannot tweak & optimize machine learning algorithm (neural networks, svm, etc.).

Therefore you need well tested X86 or neural network AI chip solution. It is more about software than hardware.

It is more about software than hardware.

Except you need the right hardware to run the software, and this is an Arduino forum.

Grumpy_Mike:
Except you need the right hardware to run the software, and this is an Arduino forum.

Except that he will use Arduino for all sensors and instead of shitty shield he will use the right solution for audio analyzing. You cannot solve this with Arduino & shield. You need extra board with Kendryte chip or X86. For example you can use LattePanda (Intel x86 with Arduino co-processor). So please, stop this nonsense.

So please, stop this nonsense.

I could say the same thing to you. This is an Arduino forum, what about that are you not understanding?

If you think your posts help the OP get a solution to his problem you are sadly mistaken.

Remember he said:-

I have an idea for our project in school.

So you think a school kid is going to have the skills to pull this off?
Yes maybe in about four years he might, but it is going to take one hell of a lot more hand holding that you give in your answer.

You cannot solve this with Arduino & shield.

Yes I know. But this is the sort of advice that has been offered here in the past and there might be some results which could be “good enough” for the OP’s requirements, which at the moment, you will have to accept are not very precise.

This is why I said about the shield

But I don't hold out much hope

cyberluk:
You can do it, but not with Arduino. For voice recognition to work like this, you need more advanced chip like Kendryte or X86 Linux / Android smart watch with GRT C++ framework (Github: GRT, spoiler: I help with this project). Or Google Tensorflow.

I am sound engineer and music producer. The waveform complexity is not a problem. Problem is you need better computation engine and Arduino voice recognition shield is not that. Because of: low number of samples in a dataset, optimized for voice (not good frequency spectrum), cannot tweak & optimize machine learning algorithm (neural networks, svm, etc.).

Therefore you need well tested X86 or neural network AI chip solution. It is more about software than hardware.

Basically that level of AI isn't worth it for the problem to be solved, use multiple bins labelled "PAPER" and "OTHER". They you can use the AI processors already (hopefully) in the skulls of the users of the bin(s).

I run makerspace in Czechia. If you go to makerspace, and you have any age, people like me will help. I'm 32 and some kids can make this stuff even on basic school.

Just be curious and you can do anything. Don't let it stop you!

And even if you want to be grumpy, I don't care, because this is Arduino forum and other people will look here for a solution for their similar task and when I look for something like that, on this forum, I want to see also other directions. It depends on time, it changes over time as you age is changing. Anyone will choose what is the best for him.

AI is only marketing buzz, it's called machine learning, not AI. You can use logical regression. You can search internet and there is a lot of GUI stuff and websites, where you can train your data model, for example by uploading mp3 on website like you upload photos on Facebook, and you don't need to write any code.

For machine learning, the point is: you don't write any code. You only provide examples and tweak some settings like you tweak servo angle on Arduino. Been there, done that.

Just to round things off I came across this yesterday, which might be of interest.